Tuesday, 24 January 2017
4E (Washington State Convention Center )
Climate models used for long-term predictions and projections are currently unable to resolve the cloud-scale physical processes responsible for rainfall due to computational limitations on model grid spacing. These processes are instead represented using moist convection parameterizations. Deficiencies in these parameterizations lead to errors in climate model simulations of rainfall patterns, which degrade their predictive skill for phenomena like El Niño and climate change. Convection parameterizations have typically been derived from physical assumptions, making only limited use of observational data to constrain them. In particular, predictive or causal relationships present in the data were not explicitly considered due to limited data availability. Since the launch of the Tropical Rainfall Measurement Mission (TRMM) in 1997, and the follow-up Global Precipitation Measurement (GPM) mission in 2014, a large high-resolution 3-dimensional dataset of rain rate measurements has become available. We apply “Big Data” analysis techniques to this rain rate dataset in conjunction with contemporaneous atmospheric flow data from NASA MERRA-2 reanalyses. We use state-of-the-art spatio-temporal statistical models, focusing on rainfall in the tropical Pacific Ocean, with different models being fitted to different rain types (i.e., shallow convective, deep convective, and stratiform). Empirical Orthogonal Function (EOF) analysis is used in the vertical dimension to isolate the dominant patterns of column variability in atmospheric state variables (temperature, humidity, and wind shear). Initial results in fitting a logistic regression model exhibit predictive empirical relationships between the dominant vertical modes of atmospheric variability in reanalyses and the different types of rain. The role of traditional predictors of moist convection, i.e., temperature and humidity profiles, as well as the role of non-traditional predictors, such as vertical wind shear, will be discussed.
![](images/flag.png)
![](images/blue_ribbon.png)